Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Mechanistic Interpretability Benchmark

Benchmarks

Task NameDataset NameSOTA ResultTrend
Indirect Object IdentificationMechanistic Interpretability Benchmark (MIB) Indirect Object Identification (IOI) (standard)
CMD0
12
Multiple-Choice Question AnsweringMechanistic Interpretability Benchmark (MIB) MCQA (standard)
CMD0.04
9
Circuit localizationMechanistic Interpretability Benchmark (MIB)
IOI83
9
Showing 3 of 3 rows